[NumPy-Tickets] [NumPy] #2023: float128 does not convert correctly from strings

[NumPy Trac]

#2023: float128 does not convert correctly from strings
----------------------+-----------------------------------------------------
 Reporter:  parejkoj  |       Owner:  somebody   
     Type:  defect    |      Status:  new        
 Priority:  high      |   Milestone:  Unscheduled
Component:  Other     |     Version:  1.6.1      
 Keywords:            |  
----------------------+-----------------------------------------------------
 Converting strings to float128/dtype('f16') does not produce the desired
 precision, but rather converts to float64 first. I'm using OS X 10.6.8,
 numpy 1.6.1, python 2.6.7, compiled under Fink.

 Example:

 {{{
 a=np.float128('0.12345678901234567890')
 b=np.float64('0.12345678901234567890')
 repr(a)
 repr(b)
 a==b
 }}}

 produces:

 {{{
 '0.12345678901234567737'
 '0.12345678901234568'
 True
 }}}

 When it should produce:

 {{{
 '0.12345678901234567890'
 '0.12345678901234568'
 False
 }}}

 I couldn't find another bug report about this, but this email thread
 describes the same problem:

 http://comments.gmane.org/gmane.comp.python.numeric.general/46380

 Whereas this bug report from two years ago seems to show correct behavior:

 http://projects.scipy.org/numpy/ticket/1311

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/2023>
NumPy <http://projects.scipy.org/numpy>
My example project